Key Takeaways:
- UC Berkeley researchers found that AI did not free up workers’ time but instead intensified their workload, which means leaders must decide in advance what saved time will be redirected toward.
- “Time saved” is the weakest measure of AI ROI, and the real question is whether that time gets redirected to higher-value work like better decisions, client relationships, and strategic projects.
- AI amplifies whatever system it is plugged into, so messy workflows produce confusion faster while clear workflows compress cycles.
- The antidote is workflow optimization before AI integration: apply Lean and Theory of Constraints to find what matters, then use AI to accelerate it.
A team I worked with last year celebrated saving 12 hours per week with AI tools, and then they filled those 12 hours with more meetings. The tool worked exactly as designed, but the organization never decided what “better” was supposed to look like once the time freed up.
This pattern is not unusual. UC Berkeley Haas researchers spent eight months embedded in a 200-person tech company and found that workers using generative AI did not work less. They worked faster, took on broader projects, and often extended into more hours voluntarily (Ranganathan & Ye, Harvard Business Review, February 2026). ManpowerGroup’s 2026 Global Talent Barometer (n=13,918 workers across 19 countries) showed that AI usage jumped 13% in 2025 while confidence in using those tools dropped 18%, suggesting people were using AI because they felt they had to, not because they believed they were using it well.
This is the efficiency trap, and it catches well-intentioned engineering firms more often than any other adoption failure mode I see.
The Four Walls of the Trap
After working with engineering firms on AI implementation through dozens of projects, I have identified four distinct problems that combine to create the efficiency trap, and each one masquerades as a simple operational issue while actually running much deeper.
Wall 1: The Measurement Problem
Your team saved 10 hours this week, and nobody can connect that to the bottom line. Time saved is the easiest metric to track and the least meaningful one to optimize. It tells you speed changed, and it says nothing about whether the work being done faster actually moves a business metric.
In my practice, we use a different measurement framework. Instead of asking how many hours AI saved, we ask what the team did with the recovered time and whether that activity created business value. If the answer is “they did more of the same work,” the AI investment created volume without creating impact.
Wall 2: The Trust Tax
Every AI output requires some level of verification, and that verification takes time. When the time required to verify is unpredictable, the net benefit of AI becomes uncertain, and uncertainty quietly kills adoption from the inside.
For engineering firms where deliverables carry professional liability, the trust tax is real and significant. A calculation that is 95% correct still requires a PE to find and fix the other 5%, and sometimes finding that 5% takes longer than doing the calculation from scratch. I have watched three different firms in the past six months stall on AI adoption for exactly this reason.
The solution is to focus AI on workflows where the verification burden is manageable: first-draft documentation, internal communications, formatting, scheduling, and knowledge retrieval. These are tasks where “95% correct” requires a quick edit rather than a full technical review.
Wall 3: Process and Context Debt
AI amplifies whatever system it is plugged into. If your workflows are clear, documented, and logically structured, AI compresses cycle times and produces useful output. If your workflows are messy, undocumented, and dependent on tribal knowledge, AI produces confusion at a higher speed and with more confidence than any human ever could.
If you cannot explain the process to a human intern in three sentences, AI will generate a polished-looking but fundamentally confused version of it.
This is why, in our methodology, we apply Lean waste elimination and Theory of Constraints analysis before introducing AI into any workflow. Think of it like troubleshooting a process system: you would never increase throughput on a line with a known constraint downstream. You would identify the constraint, resolve it, and then increase flow. AI integration follows the same engineering logic.
We applied this approach recently with an industrial distributor and found that two of the five workflows they wanted to automate did not need AI at all. They needed the unnecessary steps removed first, and then the remaining work became simple enough that basic automation handled it. The three workflows that did benefit from AI performed dramatically better because the underlying process was clean.
Wall 4: The Incentive Mismatch
This is the wall that leaders talk about least and employees feel most acutely: if saving time just means being expected to produce more output at the same pace, there is no rational reason for anyone to reveal their efficiency gains.
The two paths are clear. Path one says “you saved time, now produce more at the same pace.” Path two says “you saved time, now focus that time on the work that creates the most value.” If an organization consistently takes path one, employees will protect themselves by underusing tools or simply not mentioning the gains they have already achieved.
This behavior makes rational sense from the employee’s perspective. People do not resist becoming more productive. They resist working in an environment where improvement becomes punishment, and if every efficiency gain gets absorbed into higher expectations without higher autonomy, that is exactly what the environment teaches them.
The Antidote: Redirect, Do Not Just Accelerate
An engineering leader who uses AI to free up Tuesday afternoon for a client visit or a strategic conversation with a key hire creates more lasting value than one who uses the same time to produce three additional reports that nobody reads. Productivity measured in volume is only valuable if the direction is correct and the output matters.
Here is what we recommend to every engineering firm we work with:
Before implementing AI on any workflow, answer two questions. First, identify the actual constraint in this workflow using Theory of Constraints thinking. Second, identify the waste that exists around the constraint using Lean analysis.
Then, before deploying AI, make two decisions. Decide what the team will do with the recovered hours once this task takes less time, and decide how you will measure whether the redirected time creates business value beyond “hours saved.”
Without those two decisions made in advance, you will get the efficiency trap every time.
In one client engagement, we identified that the primary bottleneck was in proposal development rather than project execution. Engineers were spending 30 to 40 percent of their time on proposals rather than billable project work. By applying Lean analysis to the proposal workflow, eliminating redundant steps, and then integrating AI for first-draft generation, we cut proposal development time by 70%. The critical decision was what happened next: the recovered time was explicitly redirected to client development and project delivery rather than producing more proposals at the same pace.
That is the difference between using AI for speed and using AI for something that actually changes the trajectory of the business.
The Complete Picture
AI does not inherently make engineering firms more productive, because what it actually does is make them faster, and faster is only valuable if the direction is correct and the destination matters.
The firms that break out of the efficiency trap are the ones that optimize workflows before adding AI so they know the constraint, decide in advance what saved time will be used for, measure business outcomes rather than hours saved, and reward teams for redirecting time toward significance rather than just absorbing it into higher volume expectations.
That is the core of what we deliver through our Crawl-Walk-Run-Sprint methodology: the workflow clarity and strategic direction that make AI tools actually move the business forward rather than just making the existing treadmill spin faster.
Your next step: The free Engineering Acceleration AI Roadmap helps you identify which workflows to optimize first, estimated weekly time savings for your team size, and a prioritized action plan for where AI creates real business value rather than just speed. It takes about 10 minutes and gives you the constraint-first clarity this post describes.
